I'm not sure I would agree that the spec should say what doesn't happen, as opposed what does happen.
so for #1 - the spec doesn't say, so it doesn't happen.
For #2: Another way to look at this is that the delegation CSRs determine which mode will handle the trap.
Only then are the xTVAL, xCAUSE, xEPC, xIP, etc CSRs written, so only one of them is written -
nothing gets transferred from one to the other.
For #3, the spec is pretty clear about which CSRs must have specific reset values (as well as GPRs).
There aren't very many... and again, if the spec doesn't say anything about a reset value, then no reset value can be assumed
(and implementations can set them to anything they want)
I will admit that there is a little ambiguity about the phrase. "at the beginning of the trap handler, sscratch is swapped..." in sec 3.1.13 and 4.1.6
because that could be interpreted to mean *before* the first instruction of the trap handler, as opposed to instructions *inside* the trap handler,
This is made more ambiguous by "is swapped" rather than "software should swap" .
It isn't a requirement that they be swapped in HW or SW (though you won't be able to safely RET if you don't).